ons 2003-02-12 klockan 20.07 skrev Duane Wessels: > CARP > > squid-2.5 CARP configuration/implementation is different > than squid-2.x head. For example, 2.5 uses 'carp-load-factor=' > but 2.6 uses 'weight='
The CARP update is probably a bit too big for 2.5.. it is a major rewrite of the CARP implementation to hopefully be somewhat closer to the (expired) I-D and easier to use. http://devel.squid-cache.org/old_projects.html#carp The patch should quite easily to 2.5 I think, but there probably have been some bugfixes in the main tree after this got merged so some more work is required.. probably better to copy the implementation from Squid-3 if this is to be done. Not sure if anyone is actually using the Squid "CARP".. and the name "CARP" is somewhat of a misnomer as it is only using a small fraction of the CARP specification in slight odd manners.. > annoying bug in Number of fields per header distribution > > Bug counts 0 fields per header when calling httpHeaderClean() > for new headers, before they are even used: > > id #flds count %total > 1 0 13363088 49.75 > 2 1 18216 0.07 > 3 2 979 0.00 A fix for this is welcome. > ICP/HTCP treated differently. In some places, ICP and HTCP are > treated as equivalent. For example, in "Peer Selection Algorithm" > stats, "ICP" also counts HTCP. But in other places, HTCP is not > also counted. For example: > > - "Number of ICP messages received" > - "Number of ICP messages sent" > - "Cache Client List" stats > - "List of Unknown sites sending ICP messages" > - ...cacheClientIcpRequests.A.B.C.D in SNMP Note: HTCP is disabled by default in 2.5.. > syscalls.disk.unlinks confusion > > The counter is incremented for: > - unlinks performed by main squid process > - unlinks performed by DISKD processes > The counter is not incremented for: > - unlinks performed by AUFS threads > - unlinks performed by the unlinkd process Fixing this could be acceptable for 2.5 I think. > cachemgr "Cache Utilization" page > > this page gives too much detail to be useful to most > users, and the name sucks. it should be removed. You are welcome to do whatever you like with this page for Squid-3, but I think it should stay in 2.5. > in storeCheckCachable(), no.release_request always 0 > > because storeReleaseRequest() clears ENTRY_CACHABLE, and > such objects get counted as no.not_entry_cachable instead Not sure I understand this one without digging into the code.. is it about some "cachemgr" statistics field? > FETCHES under "Cache Client List" > > the FETCHES number counts requests forwarded for any reason > (ICP, HTCP, Cache Digests, default parent). > Following the FETCHES count is a percentage, based on the > number of ICP replies. This percentage may be larger > than 100. Just remove the percentage, or only count > fetches due to ICP? Remove the percentage, or perhaps make it a percentage of all cache misses. > DISKD Q1, Q2 confusion > > documentation does not match the implementation. Tried to > convince henrik of this before, but failed. You can try again. I am not strict about this, and as I am not using diskd I can only speak based on my experiences from aufs with similar queue limit values.. > Redirectors recieve URIs with whitespace > > if "uri_whitespace allow" then redirectors may receive > URIs with whitespace and not properly parse them. This should be fixed I think. > IpcacheStats and FqdncacheStats > > I removed some members of IpcacheStats and FqdncacheStats > in squid-2 HEAD, but did not MFC to squid-2.5 (or squid-3) Everything committed in Squid-2 HEAD still gets merged into Squid-3 after a while (as said in the notice about Squid-2.6 killed). Is there actually any reason to make this change in 2.5? Considering that 2.5 is a STABLE branch which should not be used for further developments.. i.e. does it change anything except for internal code cleanup? -- Henrik Nordstrom <[EMAIL PROTECTED]> MARA Systems AB, Sweden
